This piece first appeared in German. The English version is a rewrite rather than a line-by-line translation, and the German original stays online in the archive: Geht uns die Arbeit aus? Warum die Job-Debatte über AI in die Irre führt.

Firms in the New York Fed's district report sharply more use of AI and very seldom report a layoff connected to it. What they describe instead is retraining, redistributed tasks and targeted upskilling. That is the trend today.

Set that beside the number everyone quotes. Goldman Sachs economists estimated in 2023 that the equivalent of around 300 million full-time jobs worldwide is exposed to automation. The figure describes how much work gets reshaped, and says nothing about how much disappears. Both things are true at once, and the distance between them is where the useful argument lives.

Every wave began with the logic of loss

The history of technology is also a history of feeling. Online shops were going to end retail, robots on factory floors were going to produce mass unemployment, and computers in offices were going to make secretarial work extinct. Each time the starting point was subtraction, and each time what got missed was everything on the other side of the ledger: new occupations, different processes, higher expectations.

Industrialization reduced manual labor and entire fields exploded around it, in engineering, logistics and energy. Computerization thinned out typing pools while software, networks and digital services grew at a pace nobody had budgeted for. Automation moved standard processes into machines and turned quality assurance, process design and robotics into careers. Every wave destroyed routine and produced better-paid, more demanding work.

AI is the next iteration of that story, and it takes the routine and leaves room for work that needs context, judgment and communication, which is where human strengths sit. The question is which work comes into being, and who is competent to take it.

What the data actually shows

Nobody serious is claiming the exposure is small. The Goldman estimate is a large number and it describes scale. On one side sits real pull on productivity and cost, and on the other sits real responsibility for qualification and for the transitions people have to make.

PwC found considerably higher productivity growth in AI-intensive industries than in more traditional sectors. That suggests the economic benefit reaches actual tasks and wages, in the organizations that manage the turn.

From inside firms the picture holds. In a large OECD survey, employers and employees both assessed AI use largely positively, particularly on performance and working conditions, while noting that job risks need watching. The opportunities are tangible and the risks are governable, which is a different sentence from the one the headlines write.

The same pattern shows up in five professions

In consulting, research, benchmarking and first syntheses can be prepared in hours, where they used to take days. The value forms after the data rush. Which numbers hold, which are irrelevant, which hypothesis survives contact with market logic, cost of capital and culture. AI supplies material and consultants supply meaning.

Tax work gets faster, cleaner and more reproducible, and clients do not pay for PDFs. What they pay for is orientation: the room to maneuver, the weighing of risk, the explanation they can act on. In law, document review, clause search and first lines of argument all accelerate, and no client trusts a black-box answer. What is expected is the placement in law, the weighting of precedent, the negotiating posture.

Audit sees fewer mechanical reconciliations and more anomaly detection, which shifts the learning curve: juniors learn earlier why something does not fit, and no longer only that it does not. In investment banking, prospectus drafts, research syntheses and scenario calculations arrive in record time, and the decisive five percent, the deal structure, the timing, the negotiation, stays human.

Education is the lever that is already moving

The large initiatives have started. Cisco has committed to training a million more people in the United States in AI and digital skills over four years, through its Networking Academy and Learn with Cisco programs. The signal to other companies is that competence scales as readily as redundancy does.

Partnerships between technology and government are following. In July 2025 Virginia introduced its AI Career Launch Pad with Google, offering scholarships for certificate programs that open access to fast-growing fields. A labor market is not only disrupted; it also gets designed.

Christopher Pissarides, the Nobel laureate at the London School of Economics, supports the same direction in the review of work and wellbeing he led. AI can raise the quality of work where its introduction and the participation around it are organized intelligently. Innovation with employee participation beats technology imposed over people. The finding has empirical backing. No firm should read it as a template.

The World Economic Forum makes the adjacent point. AI can drive value, productivity and good work at the same time, provided organizations redefine roles and invest in new capabilities. The argument is about capability plus technology deciding who wins.

The new roles sit at the interfaces

An AI ethics manager translates legal, cultural and reputational risk into guardrails people can work with. Data translators close the gap between data teams and the business, and decide in practice whether analytics has any effect. The prompt or pattern architect designs input and process patterns that hold up inside complex agent workflows. Somebody has to build the control points where human judgment calibrates the machine, and that is the human-in-the-loop designer.

Those profiles line up with the trio of trainers, explainers and sustainers that Wilson, Daugherty and Morini-Bianzino described in 2017: people who instruct AI, explain it and run it responsibly. Work of that kind sits at the core of knowledge work now.

Which makes this a talent debate

Work is not disappearing, it is moving, and the criteria for talent move with it. Who builds the cleanest spreadsheet counts for less. Transfer counts for more, between industries, datasets and cultures. So does staying critical when a model flashes an F1 score of 0.98, and communicating so that a board, a supervisory body or a credit committee can carry a decision.

The facts are clearer than the debate suggests. Few AI-related layoffs so far, and a great deal of retraining. Productivity gains show up where AI is seriously integrated into processes, and the assessments from both sides of the employment relationship are broadly positive. And a large exposure of tasks, which calls for design work.

Companies that want to act on this have four things to do. Posture comes before tools, because AI replaces routine and never responsibility, and the learning paths have to build thinking, with case shadowing, simulations and explain-your-reasoning as standard. Mentoring becomes obligatory, so that seniority is measured by how well somebody explains. And the new roles get defined properly, with metrics, budget and a career path. For policymakers the list is shorter: scale certificate and scholarship programs, make data and AI competence binding in curricula from school onward, and treat the productivity gap as something to close.

A personal observation: in conversations with juniors I see two patterns. Some love the tools, work at tremendous speed, and take the results without examining them, which is comfortable and dangerous, and it is the Midjourney mentality. Others use AI as a sparring partner, putting hypotheses against it, testing its limits, documenting uncertainty. They are not always the louder group, and they are the more effective one. The future belongs to the fast thinkers rather than the fast.

Sources

Grouped by the section they support.

Opening

  • Abel, Deitz, Emanuel, Hyman and Montalbano (2025), Are Businesses Scaling Back Hiring Due to AI?, Liberty Street Economics, Federal Reserve Bank of New York, 4 September 2025, newyorkfed.org. Supports rising AI use among firms in the New York–Northern New Jersey region, very few AI-related layoffs, and retraining as the most common response. The 2026 follow-up confirms the pattern: Abel, Deitz, Emanuel and Montalbano (2026), Businesses Are Using AI to Transform Work, Not Cut Jobs, 1 September 2026, newyorkfed.org.
  • Goldman Sachs (2023), Generative AI could raise global GDP by 7%, 5 April 2023, goldmansachs.com. Supports the estimate that generative AI could expose the equivalent of 300 million full-time jobs to automation.

What the data actually shows

  • PwC (2025), The Fearless Future: 2025 Global AI Jobs Barometer, press release, 3 June 2025, pwc.com. Supports considerably higher productivity growth in the industries most exposed to AI.
  • Lane, Williams and Broecke (2023), The impact of AI on the workplace: Main findings from the OECD AI surveys of employers and workers, OECD Social, Employment and Migration Working Papers No. 288, oecd.org. Supports the largely positive assessment by employers and workers of AI's effect on performance and working conditions, alongside concerns about job loss.

The same pattern shows up in five professions

  • No external source. The observations on consulting, tax, law, audit and investment banking are the author's own.

Education is the lever that is already moving

  • Katsoudas (2025), Advancing U.S. AI Leadership: Cisco to Skill 1 Million People, Cisco, 4 September 2025, blogs.cisco.com. Supports the commitment to train one million more people in the United States in AI and digital skills over four years, through Learn with Cisco and the Networking Academy.
  • Office of the Governor of Virginia (2025), Governor Glenn Youngkin Unveils New "Virginia Has Jobs" AI Career Launch Pad in Partnership with Google, 17 July 2025, via govirginiaregion8.org. Supports scholarships for Google AI Essentials courses and Google Career Certificates.
  • Pissarides, Thomas et al. (2025), The Pissarides Review into the Future of Work and Wellbeing: Final Report, Institute for the Future of Work, 26 January 2025, ifow.org. Supports the finding that good outcomes from automation must be consciously shaped, and that information, consultation and investment in training enable purposeful technology adoption.
  • World Economic Forum (2024), Leveraging Generative AI for Job Augmentation and Workforce Productivity, 25 November 2024, weforum.org. Supports generative AI improving job quality and productivity where organizations align strategy with their workforce.

The new roles sit at the interfaces

  • Wilson, Daugherty and Morini-Bianzino (2017), The Jobs That Artificial Intelligence Will Create, MIT Sloan Management Review, 1 July 2017, sloanreview.mit.edu. Supports the categories of trainers, explainers and sustainers. The four role profiles are the author's own.

Which makes this a talent debate

  • No external source. The recommendations draw on the author's professional experience.
The link has been copied!